2634

SHACKELFORD ET AL.

selected using the RSQUARE procedure (SAS, 1988), w h i c h s e l e c t s t h e s i n g l e b e s t ( h i g h e s t R 2 ) e q u a t i o with a given number of variables. Thus, the RSQUARE technique differs from STEPWISE tech- niques in that the variables selected for higher-order equations do not depend on the variables used in lower-order equations. Equations were evaluated with n

2 respect to R , the

p C statistic (Mallows,

1973), and

residual standard

deviation ( RSD). To

more fully

evaluate the ability of carcass cutability, RPYD

image analysis to predict was regressed against yield

grade, and RPWT was predicted predicted RPYD times HCW.

by

multiplying

Following collection of the aforementioned images, each steak was repositioned, and a second image was captured. The second image was not used in develop- ment or validation of regression equations. Those images were used to test the repeatability of the image analysis process. Retail product yield and SLA were predicted using the optimal equations described below, and repeatability of those prediction estimates was calculated using the VARCOMP procedure of SAS. For

Results

Simple statistics of carcass traits and independent and dependent variables are presented in Table 2. On average, the carcasses used in the present experiment were lighter, had less fat thickness, larger longissimus area, lower yield grades, and lower marbling scores than are typical of the U.S. beef industry (Lorenzen et al., 1993; Boleman et al., 1998). This result was expected based on the high frequency of the double muscling allele in this population (Hanset, 1991; Casas et al., 1998; Wheeler et al., 1997).

Retail Product Yield. Regression equations are presented in Table 3. The single image analysis variable accounting for the greatest proportion of

variation in RPYD was PERLEAN (R best five-variable equation (Equation 15, 2

= .77). The in Table 3)

o p t i m i z e d R 2 ( . 8 8 ) , four- and five-variable C

p statistic, and equations most

RSD. The best p r e c i s e l y ( R 2 =

.91)

predicted retail

data

set. When data

and

validation data

product yield in the validation were pooled across development sets, the five-variable equation

each trait, repeatability was calculated as

(

2

carcass

+

2

error

).

2

carcass

/

accounted for more (89 RPYD than did yield

vs 77%) of the

variation

grade (Figure

2).

in

Total lean area, pixels

LEAN

49,408.0

6,794.8

14

33,869.0

65,742.0

Total fat area, pixels

FAT

28,911.0

8,176.3

28

12,036.0

44,219.0

Total steak area, pixels

TOTAL

89,333.0

9,604.8

11

68,551.0

111,731.0

Area of largest lean piece, pixels

EYEPIECE

39,990.0

7,992.7

20

22,413.0

60,884.0

Mean red intensity of EYEPIECE

RED

157.2

11.0

7

126.9

178.2

Mean green intensity of EYEPIECE

GREEN

37.6

9.6

26

20.5

65.7

Mean blue intensity of EYEPIECE

BLUE

44.1

9.5

21

27.8

71.5

Mean density of EYEPIECE

DENSITY

79.7

9.4

12

61.2

104.7

Number of holes in EYEPIECE

NUMHOLES

428.4

141.3

33

194.0

858.0

Area of holes in EYEPIECE, pixels

HOLEAREA

3,513.6

1,551.7

44

980.0

8,841.0

Adjusted fat thickness, mm

—

5.8

3.3

57

1.3

12.7

Carcass longissimus area, cm^{2 }

—

85.1

12.7

15

56.8

114.8

Retail product yield, %

RPYD

72.9

6.9

10

61.4

89.1

Retail product weight, kg

RPWT

203.1

29.5

15

146.2

259.9

Steak longissimus area, cm^{2 }

SLA

83.9

12.8

15

58.7

109.7

Carcass traits Hot carcass weight, kg Actual fat thickness, mm

HCW —

295.8 6.4

34.2

12 55

222.9 1.3

374.5

3.5

15.2

Kidney, pelvic, and heart fat, % USDA yield grade Lean color score Lean maturity score^{a }Skeletal maturity score Overall maturity score Marbling score^{b }

a a

— — — — — — —

2.5 1.8 3.0 155.2 173.5 164.5 372.0

.9 .9 .7 13.5 21.3 13.9 57.2

36 50 23 9 12 8 15

.5 .4 1.0 130.0 130.0 135.0 250.0

4.0 3.6 5.0 190.0 220.0 190.0 490.0

HOLEAREA), %

PERHOLE

7.6

2.8

37

100 LEAN/(LEAN + FAT), %

PERLEAN

63.4

8.3

13

3.2

15.2

45.2

84.5

Table 2. Simple statistics of carcass traits and independent and dependent variables (n = 66)

Variable

Abbreviation

Mean

SD

CV

Minimum

Maximum

Independent variables^{c }

100 HOLEAREA/(EYEPIECE +

# Dependent variables

a b c

100 = A^{0}; 200 = B^{0}. 200 = Traces^{0}; 300 = Slight^{0}; 400 = Small^{0}; 500 = Modest^{0}. Hot carcass weight (listed above with carcass traits) was included as a potential independent variable.